AI for resource-efficient circular fashion

Our principal objective is to identify, design, and implement data-centric AI solutions tailored specifically for the second-hand textile industry, emphasizing automating the textile recycling sorting process. We aim to construct an extensive, open dataset of used clothing and use this information to train ML models to streamline and optimize the textile recycling process.

Introduction

Textiles are the fourth highest pressure category in the EU regarding primary raw materials and water (after food, housing, and transport) and fifth for GHG emissions (EEA). Europeans consume about 11 million tonnes of textiles annually; today, only 2.8 million tonnes are collected for reuse or recycling. The same is true globally, where only 20% of the yearly textile consumption is collected, whereas 80% is incinerated or ends up in landfills. To address this, a new European Waste Directive has been put forward, which states that by January 1, 2025, EU member states are obliged to have systems in place for separate collection of textiles, which will drastically increase the collected textile volumes and the need for textile sorting for both reuse and recycling. In Sweden alone, 75,000 tonnes of textiles per year are discarded and become residual waste and input material for central heat power plants.

The Role of AI and Datasets

High-quality datasets for used clothes do not exist. Data is the key to building robust AI models. We are collecting data for 30,000 clothes with three images each, annotated by expert sorters to identify brand, damage, pattern, and optimal use, among other attributes. The data will be made public at the end of the project to enable efficient automated solutions in the second-hand clothing industry.

Our solution

Our open dataset will be released under permissive licenses to allow broad use contributing significantly to circularity for textiles and more efficient sorting processes. An updated version of the dataset is available for download on Hugging Face. One of the unique challenges in the second-hand clothing industry is that many (non-profit) organizations have untrained volunteers sorting clothes daily. Extensive training is costly and time-consuming. We hope our AI-powered automated decision support tool can help solve this problem. Our ultimate goal is to considerably extend the lifespan of clothing, thereby reducing the need for new production. 

Partners

The project involves two well-established innovation environments, Wargön Innovation, and RISE. We are fortunate to partner with various partners in the second-hand clothing industry including the Red Cross, Myrorna, BjörkåFrihet, Texaid, MiniKit, and ShareTex.